REMAP Periop: a randomised, embedded, multifactorial adaptive platform trial protocol for perioperative medicine to determine the optimal enhanced recovery pathway components in complex abdominal surgery patients within a US healthcare system

Trial design

Our protocol was designed according to the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) guidelines, which are addressed in full in the SPIRIT Checklist Appendix (online supplemental appendix 1). This is a randomised, open-label, adaptive clinical trial. Here, we present the REMAP Periop core protocol, which specifies standardised trial elements that are applicable to all trial domains and serves as a guide for any future interventions or domains to be added (online supplemental appendix 2).12 13 REMAP Periop uses a perpetual, adaptive platform design with no maximum sample size. The trial is multifactorial in nature, allowing for patients to be randomised to one of several intervention arms within one of multiple possible domains simultaneously (figure 1). Thus, the core protocol allows for aggregation of the treatment response across different simultaneously investigated domains and the multifactorial evaluation of synergistic or antagonistic combinations. The efficacy of an intervention arm within a given domain will be evaluated at regular interim analyses using Bayesian statistical analysis. These analyses will also be used to guide prespecified statistical triggers and response adaptive randomisation (RAR) when employed within any given domain. At the beginning of the trial, participants will have an equal probability of being allocated to any given intervention within a domain (ie, simple 1:1 randomisation), with RAR guiding changes to allocation ratios after interim analyses when applicable. Interim analyses are scheduled to be performed quarterly. Domain-specific appendices are described within this core protocol. Trial-related patient data will be collected from the electronic medical record (EMR) for 30 days postoperatively to assess primary and secondary endpoints. Our planned trial start date is 15 May 2023.

Figure 1

Randomisation schema for the original two REMAP Periop domains, PONV prophylaxis and regional/neuraxial analgesia. ERP, enhanced recovery pathway; PONV, postoperative nausea/vomiting.

Trial embedding

This trial employs the use of embedded software designed by our institution’s clinical analytics team that interfaces with both the study investigation team and the EMR for ease of recruitment, enrolment, treatment allocation and data collection. Patients are screened as eligible for study inclusion by the software if they have one of the five eligible ERP abdominal surgery PowerPlans (table 1), which are ordered by the surgery team prior to the day of surgery if the patient meets all other inclusion criteria. PowerPlans are order sets that are placed in our EMR, Cerner PowerChart, by a patient’s provider in advance of their surgery and are activated at the time of surgery in the preoperative area. Once a patient is identified as eligible for study enrolment by the embedded software, a randomisation request is automatically triggered via a representational state transfer (RESTful) Java web service coded with the randomisation scheme described above. After randomisation, the web service replies to the EMR with coding for the allocated interventions. These orders are then available in the patient’s chart for approval and signature by the patient’s clinical team (figure 2). Though PowerPlans are ordered in advance of an elective surgery and trial interventions are allocated at that time, they are typically not opened and activated by the patient’s surgical team until they arrive in the preoperative area. The web service also codes a record of the randomisation to the trial database. A trial dashboard for automatic extraction of prespecified trial variables and endpoints was also created.

Figure 2

Workflow demonstrating interface between trial web service and electronic medical record for recruitment, randomisation and allocation, and delivery of allocated interventions back to electronic medical record for initiation by patient’s clinical team. CA, Clinical Analytics.

Study population and eligibility criteria

This study population includes patients undergoing complex abdominal surgery as part of an ERP at three hospitals within a multihospital academic institution. Patients will be identified as being an ERP complex abdominal surgery patient by having one of five PowerPlans ordered by their treating physician prior to the day of surgery. They must also meet all additional inclusion criteria and none of the exclusion criteria detailed in table 1.

PONV prophylaxis domain

ERAS consensus guidelines suggest PONV prophylaxis with at least two antiemetic medications for standard-risk patients with the addition of further medications dependent on the number of patient risk factors present.6 In order to universally treat all risk profiles, patients at our institution presently receive a three-drug regimen with the possible addition of various other prophylactic medications, dependent on patient risk factor assessment by the anaesthesiologist, patient and anaesthesiologist preference, and hospital practice patterns. The current standardised PONV prophylactic regimen for all patients undergoing abdominal surgery as part of an ERP at our institution is perphenazine 8 mg administered orally preoperatively, dexamethasone 4–5 mg intravenously at induction of anaesthesia and ondansetron 4 mg intravenously before emergence from anaesthesia. Nearly all patients receive this regimen regardless of PONV risk factor assessment due to low medication risk profiles and ease of implementation without formal risk factor assessment.

Aprepitant, a neurokinin-1 inhibitor commonly used for chemotherapy-induced nausea, has shown promising results in improving rates of PONV when added to a multimodal prophylactic regimen.14 Dimenhydrinate, a theophylline derivative traditionally used for motion sickness, is also a low-cost antiemetic that has been shown to be equally effective as ondansetron at preventing PONV.15 The use of additional medications to provide ‘supraoptimal’ prophylaxis, when compared with regimens similar to that of the three-drug combination used at our institution, has been shown to decrease PONV rates in an observational study of bariatric surgery patients.16 Variations of supraoptimal prophylaxis are practised presently at our institution and affiliated hospitals for selected patients.17 While concerns about the cost-effectiveness of a five-drug regimen are valid, dimenhydrinate is a notably low-cost medication and aprepitant has reduced drastically in price since becoming off-patent. Given the costliness of prolonged hospital stay associated with uncontrolled PONV, studies have demonstrated a linear increase in hospital net profit with increased PONV prophylaxis administration.18

In our PONV prophylaxis domain, patients will be randomised to either the optimal (standardised three-drug regimen) PONV prophylaxis or supraoptimal prophylaxis with the addition of dimenhydrinate and aprepitant (table 2). Patients will be randomised to one of the two arms within the PONV prophylaxis domain in a 1:1 ratio, as allocated by the randomisation schema integrated in the embedded software. The allocated medications will be recommended to and ordered by the patient’s treating physician/team. The treating physician has the capability to deviate from the trial protocol if contraindications to the allocated interventions exist. Pertinent treating physicians and provider teams were educated about trial interventions prior to the trial start date in order to maximise adherence when clinically appropriate. Full details about the PONV domain can be found in the PONV Domain-Specific Adaptive design Report (online supplemental appendix 3).

Regional/neuraxial analgesia domain

A wide variation of clinical practice regarding the use of different regional/neuraxial analgesic techniques in patients undergoing abdominal surgery exists both at our institution and others.19 The use of intrathecal morphine, quadratus lumborum (QL) or paravertebral blocks are acceptable neuraxial or regional pain management strategies for abdominal surgery and are variably used at our institution and others as components of ERPs. There is currently minimal evidence to support one of these strategies over the other in abdominal surgery patients. In our regional/neuraxial analgesia domain, patients will be randomised to one of five individual or combinations of regional and/or neuraxial techniques (table 3). All of these interventions employ a single-injection technique without continuous infusions or catheters. We chose this approach in order to improve patient postoperative mobility and to transition from intravenous to oral opioids as needed for potential early discharge. This also allows direct comparison of the initial 24 hours efficacy and promotes future adoption at institutions where catheter-based therapy and requisite follow-up may not be available. In this study, we chose to use QL blocks, a variation of a transversus abdominis plane (TAP) block, as this block has demonstrated superior efficacy in several studies when compared with classic TAP blocks.20 Proposed reasons for the improved efficacy of QL blocks include: QL blocks typically cover higher dermatomes (T4–L1 vs T10–L1), and also provide visceral, in addition to somatic, coverage.21 22 Finally, there is wider variation in TAP block technique and therefore using a standard QL approach decreases variability and allows for a more direct comparison to alternative techniques.23

Patients will be randomised to one of the five interventions within the regional/neuraxial analgesia domain in a 1:1:1:1:1 fashion initially. RAR and intervention removal based on prespecified statistical triggers will be employed to adjust allocation within the domain based on the accruing efficacy data during interim analyses, which will be conducted every 3 months of the study and are described in further detail in the Regional/Neuraxial Domain-Specific Adaptive Design Report (online supplemental appendix 4). The allocated intervention will be approved by and administered by the patient’s treating acute pain anaesthesiologist prior to the induction of anaesthesia. Again, the treating physician has the capability to deviate from the trial protocol if contraindications to the allocated interventions exist. Study personnel collaborated with study site acute pain physicians prior to and during trial enrolment to ensure maximum adherence to allocated injections. While there is limited data regarding the cost-effectiveness of varying regional/neuraxial analgesic techniques in patients undergoing abdominal surgery, the study investigators are sensitive to the increased costs associated with using combination techniques. We, therefore, have designed the domain with statistical triggers to remove combination arms based on futility compared with the better of two single agents comprising that combination, when applicable (online supplemental appendix 3).

Endpoints

The primary endpoint is HFD30. HFD30 is a composite ordinal outcome defined as the number of days alive and not in the hospital from the day of surgery up to postoperative day 30. For survivors, this outcome can be represented by the following equation:

HFD30=30–postoperative length of stay–cumulative readmission length of stay

where higher numbers represent faster recovery. All deaths up to day 30 are assigned the worst outcome level (designated as –1). HFD30 captures morbidity that a patient may suffer in the early postoperative period; patients who have PONV, for example, will often require longer hospital stays or readmission. This outcome measure indirectly quantifies cost utilisation (increased hospital days leads to increased system costs), and also functions as a patient-centred outcome (decreased time in hospital, faster return to work, etc).24 This endpoint also has the advantage of capturing patient mortality, and weighs it heavily by designating HFD30=–1 as described above.25 The Core Outcome Measures for Perioperative and Anaesthetic Care has identified mortality, hospital length of stay, and unplanned readmission as key outcomes for trials in perioperative care, and HFD30 addresses each of these outcomes in one.26 HFD30, therefore, serves as an endpoint which can capture treatment effect across a variety of different domains. This outcome will be abstracted from the EMR by obtaining data regarding the patients index postoperative hospital length of stay as well as the length of stay of any readmission within thirty days postoperatively. Within the PONV prophylaxis domain, we hypothesise that patients randomised to the supraoptimal prophylaxis arm will have increased hospital-free days compared with optimal prophylaxis. Within the regional/neuraxial analgesia domain, we hypothesise that patients receiving combinations of intrathecal morphine plus either paravertebral or QL blocks will have increased hospital-free days compared with any of these techniques used alone.

We will also evaluate domain-specific secondary endpoints for our two original domains. For the PONV prophylaxis domain, we will compare incidence rates of PONV. Since documentation of PONV incidences in the medical record varies across hospital sites and is not always consistent, PONV treatment medication administration is commonly used as a surrogate outcome measure for PONV incidence.27 This will be measured as a dichotomous outcome, denoted as ‘yes’ if any of the following rescue medications are administered to the patient within the 24 hours following surgery: dimenhydrinate, haloperidol, granisetron, ondansetron, prochlorperazine, dolasetron, droperidol, metoclopramide, palonosetron, perphenazine, promethazine, ramosetron, scopolamine or tropisetron. These data will be abstracted from the medication administration record within the EMR via our clinical dashboard.

For the regional/neuraxial analgesia domain, we will compare quantities of opioids consumed 24 hours following surgery as a secondary endpoint. Opioid quantities will be converted to oral morphine equivalents (OME) in milligrams per day, a continuous variable that can be compared among patients receiving different types of opioids. Standardised conversion factors for each opioid medication are well described and frequently used in literature regarding pain control.28 OME data will be abstracted from the medication administration record within the EMR via our clinical dashboard.

Statistical analysis

The primary analysis set will consist of all participants that are randomised to at least one intervention within each domain. The primary analysis set will apply the intention to treat principle by analysing all participants by the interventions to which they were randomised. Additionally, otherwise ineligible patients, as defined by the inclusion and exclusion criteria, may have one of the qualifying ERP PowerPlans ordered by their treatment team and thus automatically receive a set of randomised treatments per the trial’s embedded software. These patients will be reviewed, and their data excluded from analysis.

Inferences in this trial are based on a Bayesian cumulative logistic (proportional odds) model, which models HFD30 outcomes for each of the interventions within each domain. Statistical quantities, such as the posterior probability that one intervention is superior to another intervention, are used to evaluate prespecified decision rules in each domain. The default statistical decision rules with their associated posterior probabilities are summarised in table 4, and further expanded on in the Core protocol with domain-specific rules in the respective domain-specific appendix. The model incorporates the empirical outcomes that accumulate during the trial in terms of the observed HFD30 outcomes and prior knowledge in the form of weakly informative prior distributions. The model is designed to evolve throughout the duration of the platform. If new domains/interventions are added, then new parameters are added to the statistical model. The statistical model adjusts for the variation in HFD30 outcomes by age, sex, hospital site, surgery type, time epochs defined by date of surgery, randomisation within each domain, American Society of Anesthesiologist physical status and opioid tolerance. Opioid tolerance is defined as any prescription opioid for scheduled use within 3 months of surgery as evident in the EMR. A complete and detailed documentation of the statistical analysis plan is included in the Core and Domain-Specific Adaptive Design Reports, which are included as previously referenced appendices.

Treatment effects are characterised as ORs for HFD30 and PONV incidence, and mean difference in OME consumed for the regional/neuraxial analgesia domain. An OR of 1 indicates no difference in HFD30 outcomes between any given set of interventions within any given domain, for example. An OR below 1 indicates that the given intervention reduces the odds of worse outcomes compared with the alternative intervention.

Interim analyses will be performed every 3 months to evaluate the set of prespecified statistical triggers as well as update RAR proportions as employed by domain. If a statistical trigger is met, design adaptations may occur including but not limited to public disclosure of the results, removal of interventions and closure of domains. For example, if at any adaptive analysis a single intervention has at least a 99% posterior probability of being the optimal intervention within a domain, then that intervention will be deemed as superior to all other interventions in that domain. When this trigger is met, it is expected that the domain will be closed, future participants will receive the superior intervention, and a public disclosure of results will be made as determined by the trial steering committee. For both domains, statistical triggers can be met via either the platform-wide HFD30 outcome or the relevant domain-specific endpoint. In the regional/neuraxial analgesia domain, RAR is updated based on the comparative effectiveness of each intervention on HFD30. Members of the statistical analysis committee (SAC) relay any ensuing changes to randomisation probabilities to the clinical analytics team, which implements changes to the randomisation schema into the embedded web service. Though interim analyses that result in RAR are not disclosed to blinded study investigators or non-study clinicians, our open-label design is prone to bias. If a clinician notices more frequent administration of a given treatment arm due to RAR, it could in theory be deduced that that is the more effective of the treatment arms and result in performance bias regarding subsequent delivery of care. However, we feel this risk is minimised in that patients from three remote hospitals from at least five different service lines are enrolled in the study, so it would likely be difficult for any one clinician to notice these trends. Additionally, the postoperative orders are part of a prepopulated PowerPlan with defined criteria for opioid and antiemetic treatment administration, based on pain scores or presence of nausea and/or vomiting respectively; therefore, the clinicians have limited flexibility or influence on ultimate administration of treatments.

Due to the use of an adaptive platform design, clinical trial simulations are used to calculate trial operating characteristics rather than standard power calculation methods. The operating characteristics of the platform depend on the design of each domain; clinical trial simulations are included in each domain-specific appendices to characterise the cumulative probability of meeting statistical triggers as a function of sample size and treatment effect. For example, in a two-arm domain, the probability of demonstrating superiority on HFD30 exceeds 80% with a total sample size of 2500 subjects with a moderate treatment effect of an OR=0.8.

Safety monitoring plan

Our safety monitoring plan will include the use of a patient safety monitor. The patient safety monitor will be a clinician (anaesthesiologist) familiar with the treatments being used in this study who is involved in neither the clinical care of eligible patients nor the study itself. A service email account has been created where any study participants or clinical personnel treating study participants can direct concerns that will be reviewed by the patient safety monitor within one business day. These concerns can be relayed to the trial steering committee in a masked fashion (open session) and can be further reviewed with unmasked members of the study personnel (closed session with SAC) and then brought back to the trial steering committee in an executive fashion if recommendations about study termination or protocol modification are deemed necessary. An external data and safety monitoring board will not be implemented in this study as all study medications already have well-established safety profiles, are at least standard of care, and are variably employed in clinical practice at our institution at present. Antiemetics and the interventions within the regional and neuraxial analgesia domain are typically well tolerated by patients and have been well studied. The safety monitoring plan will be executed by the principal investigator.

Data management plan

All data necessary for the trial will be stored in a private database only accessible by the core research and project management team, excluding all clinicians and trial steering committee members. The core working group will be unmasked to the data. When the trial has reached an interim analysis, a core working group member will compile the necessary data from the clinical dashboard for the analysis to be completed. These data will be stored in a locked folder in Microsoft Teams behind an institutional firewall. Data transfer to the SAC will occur through Microsoft OneDrive. Only core working group members and the SAC members have access to the OneDrive folder, again behind the institutional firewall. When the interim analysis is complete, the SAC will upload the updated randomisation probabilities to the OneDrive folder. The updated randomisation probabilities will be updated in the randomisation coding, and the trial will continue until the next interim analysis.

Patient and public involvement

Patients were not involved in the design of this study; however, patients and the public will be notified when prespecified statistical triggers are reached that allow the study team to make conclusions about study interventions.

Ethics and dissemination

The core protocol and domain-specific appendices were approved by the University of Pittsburgh Institutional Review Board (19110079, 19030022). A waiver of informed consent was applied for and granted from the Institutional Review Board for this trial as patients are receiving at least standard of care interventions, there is wide variation in clinical practice in each domain, and the treating physician has the capability to deviate from the trial protocol if contraindications to the allocated interventions exist. Having this waiver of informed consent allows the study investigators to more readily enrol eligible patients, and therefore, reach conclusions about the most effective interventions more quickly. The expeditious implementation of the most advantageous perioperative therapies is to the benefit of our patients. We offer study information pamphlets for patients who request them. Trial results will be announced to the public and healthcare providers once prespecified statistical triggers of interest are reached as described in the core protocol, and the most favourable interventions will then be implemented as a standardised institutional protocol. As these domains continue to enrol and eventually conclude, the REMAP platform allows for the possibility of adding additional domains in the future. Considerations for other domains that might be included in the future include perioperative care elements such as the management of intraoperative hypotension, selection of primary anaesthetic agents and the use of other anaesthesia adjuncts (such as ketamine and lidocaine infusions).